Rename PKLookupTask to BatchPKLookup (#555)

[serhiy-bzhezytskyy]

Renames PKLookupTask → BatchPKLookup per #555, since it looks up N primary keys per task (the batch variant).

• Renamed the class and its constructors, updated the call site in LocalTaskSource, and the javac classpaths in compile.sh and responseTimeTests.py.
• Kept the reported category as "PKLookup" so nightly-chart history is unaffected, and left the historical annotations in notation.py as-is (they describe past events under the old name).
• The Python side in benchUtil.py matches on the emitted TASK: PK… line and the PKLookup category, both unchanged, so it needs no changes.

Verified: src/main/perf/*.java compiles against a built Lucene 11.

I kept the sibling classes (PointsPKLookupTask, PKLookupWithTermStateTask) as-is; happy to rename those too for family consistency if you'd like.

[serhiy-bzhezytskyy]

Are there any objections to its merger? I am ready to resolve them, just let me know. Thanks

[mikemccand]

Thank you @serhiy-bzhezytskyy! +1 to rename the other two cases as long as they are also batch-optimized?

[mikemccand]

Looks great, thank you @serhiy-bzhezytskyy! I'll merge shortly.

[serhiy-bzhezytskyy]

Thanks for merging!

On renaming the other two — I checked, and they don't meet the condition, so I'd leave their names alone:

• BatchPKLookup sorts the ids before looking them up, with the comment saying exactly why: // Sort so lookups are in term-order, optimizing for Lucene's fastest batch PK lookup use case. That's what earns the "Batch" in the name.
• PKLookupWithTermStateTask does two passes over ids in generation order — first seekExact recording a TermState per segment, then re-seeking with that TermState to resolve docids. No sort, so no term-order locality; what it measures is the TermState caching, not batching.
• PointsPKLookupTask builds an IntPoint.newExactQuery per id, also unsorted.

Neither has an Arrays.sort, so calling them "Batch…" would make the name assert an optimization the code doesn't do.

If you'd like them batched too, that's a small separate change — sort the ids in each, then rename. Worth noting it would change what those two tasks measure, so their numbers would shift and the history before/after wouldn't be comparable. I can put that up as its own PR if you think the parity is worth the discontinuity.